Robot touch explained: who is building tactile hands and sensors, and where the money is going in 2026

Robots that can feel what they grip are no longer a research sideshow. In the past six weeks Singapore's Sharpa launched a robot covered in electronic skin, Rhoda AI showed a robot retrieving objects by touch alone with its cameras blocked, and Chinese gripper maker Jodell launched a tactile five-finger hand on Oct 16. The short answer to who leads: in hands with touch, Sharpa and Korea's Wonik Robotics have the most visible products and capital; in using touch as training data, Rhoda AI and a handful of research groups are ahead.

The harder question is where the money will end up. Sensors are getting cheaper and more numerous, which usually means the margin moves elsewhere. This piece maps the players from MEKAWIRE's verified records and argues that touch is becoming a data business.

How robot touch actually works

Most robot hands today sense position through joint encoders and, at best, force at the wrist. Tactile sensing adds contact information at the fingertip or across the skin: where the object touches, how hard, and whether it is slipping. Three families dominate. Resistive and capacitive electronic skins measure pressure across a grid. Visuo-tactile sensors put a small camera behind a soft gel and read how the gel deforms, which gives very fine contact geometry. Biomimetic sensors such as the BioTac line, which Jodell now builds on with SynTouch founder Gerald E. Loeb, combine pressure, vibration and temperature in a fingertip.

The engineering trade-off is resolution against durability. A gel that deforms finely also wears; a skin that survives a factory shift often sees less. That is why the commercial question has shifted from which sensor is best to which data a hand can collect cheaply and reliably.

Who is building what

Sharpa, based in Singapore, is the most complete hardware stack in our records: its W02 hand has 21 active degrees of freedom at under 750 grams, its D01 robot carries whole-body electronic skin, and its AE01 haptic glove captures human touch data for teleoperation, all launched at IROS 2026. NVIDIA chose two 22-degree-of-freedom Sharpa Wave tactile hands for its Isaac GR00T reference humanoid on Unitree's H2 Plus body, announced on June 1.

Rhoda AI, which raised a $450m Series A in March, built its own low-cost tactile sensor and extended its video-action model to predict future touch signals, according to its Oct 13 research post. Tangent Robotics, a Columbia spin-out, raised a $4.5m pre-seed co-led by Fly Ventures and Toyota Ventures on Sept 30 for hands with light-based touch sensors aimed at connector mating and gear meshing.

In Korea, Wonik Robotics is issuing KRW 350bn of convertible preferred shares to a Korea Development Bank fund and an IMM vehicle to build a robot-hand factory in Wanju. AIDIN Robotics, a force and tactile sensor maker, raised KRW 16bn, of which KRW 13bn came from HD Hyundai Robotics for a heavy-duty five-finger hand for HD Hyundai's humanoid. Tesollo expects cumulative hand and gripper sales to pass 1,000 units by the end of 2026, according to a Robot Newspaper report the company has not confirmed. Korean materials maker Unichem's purchase of US e-skin company Loomia has stalled on a $1.5m instalment awaiting foreign-exchange approval, per its Oct 16 filing.

The evidence that touch improves robots

The research case is getting firmer. The Spark-0 preprint from Xspark AI, Tsinghua and partners trained on 1,000 hours of human touch data and lifted an egocentric humanoid benchmark from 52.45% to 71.72% success. Rhoda's sensor spans a few millinewtons to more than 40 newtons at about 500 Hz, though it disclosed no success rate. The caveat is that most gains are still measured on benchmarks, not production lines.